INTRODUCTION: The dense extracellular matrix in pancreatic ductal adenocarcinoma (PDA) is an underlying mechanism for treatment failure therefore overcoming this barrier has motivated developing small molecules and biologics which can degrade or reduce matrix components in PDA ^1-5. We aim to test the utility of dynamic contrast enhanced (DCE)-MRI to detect and predict responses to stroma-directed interventions. Specially, we compared DCE-metric (K^trans, k_ep and V_p) obtained from pharmacokinetic (PK) modeling in combination with individual- arterial input function (AIF) or a group-AIF for the ability to detect treatment effect. Corroborating imaging and immunostaining data, this study aims to provide rationale supporting choice of PK model and AIF approach which lead to quantitative imaging marker with optimal sensitivity and specificity for stroma-directed drug.
METHODS: Cells (4662-KPC) were obtain from dissected tumor in KPC mice, a genetically engineered mouse (GEM) model of PDA ⁶ as described earlier ³. An orthotopic PDA model was generated by injection of 1.25x10⁵ 4662-KPC cells into the pancreas of syngeneic mice by a surgical procedure ⁷. Mice were subjected to DCE-MRI at baseline and 24 h after one iv injection of PEGPH20 (1 mg/ kg) or vehicle (VEH). PEGPH20 (PP) is an investigational drug that degrades hyaluronan (HA) in PDA extracellular matrix ². Tumors were harvested after imaging for assessing HA level by immunostaining. MRI studies were performed on a 9.4T DirectDrive® system (Agilent) interfaced with a 12-cm gradient coil. DCE protocol to generate T₁₀ map of tumor and blood (T₁ before CA injection), DCE-series, AIF and for individual mice were described earlier ⁷. MultiHance® diluted to 10 mM of gadolinium in saline was injected in 0.2 mL via tail vein. Group-AIF was the average of 20 AIFs measured from 10 mice; each mouse contributed two AIFs, one obtained at baseline and the other after treatment. Individual AIFs were aligned by bolus-arrival time of CA before averaging. DCE series and T₁₀ maps of the tumor were fit to a PK model using the least-squares method. Three PK models were compared: Tofts (T), Modified-Tofts (M-T) and Shutter-Speed (SS) model ^8-10. Pixel-wise parametric maps of K^trans (the rate constant of transferring unit volume of contrast agent from capillaries to interstitial space, min^-1), k_ep (the rate constant from interstitial space to capillaries, min^-1), τ_i (the intracellular water life time, sec),V_e (extracellular & extravascular volume fraction =K^trans/k_ep) and V_p (vascular volume fraction) were obtained as output. Receiver Operating Characteristic (ROC) curves of K^trans were constructed in SPSS (IBM). Responses to PEGPH20 was confirmed by reduction hyaluronan level by IHC compared to VEH-treated mice ⁷.
RESULTS: While all mice were from the same strain and similar in age, individual AIFs exhibit a great difference in amplitude, leading to the highest SD at the peak point of group-AIF (Fig 1A), consistent with other report ¹¹. Variation of AIF among subjects or between pre- versus post-treatment in same mouse can result from differing physiological condition at the time of study or occasionally imperfect bolus injection of contrast agent. Pixel-wise K^trans (Fig 1B) and mean K^trans of each tumor (Fig 1C) exhibit moderately good correlation between individual versus group-AIF with Pearson coefficient r =0.69 and 0.86, respectively.
Employing individual-AIF, both SS- and T-model detected a significant increase of k_ep after PP treatment (Fig 2A), whereas only SS (but not T or M-T) model detected a significant increase in %change of K^trans between PP versus VEH group (Fig 2B). Using group-AIF, only SS model was able to detect a significant increase of k_ep and K^trans from baseline to post-PP (Fig 2C-2D) as well as an increase in %change of K^trans between PP versus VEH group (Fig 2E).
Data in Fig 2 suggest that K^trans and k_ep are suitable metric for detecting the effect of PEGPH20; meanwhile, the SS-model seems more sensitive than T or M-T model when individual or group-AIF approach is employed. We computed the ROC curves of K^trans and k_ep (Fig 3A-3B). AUC (area-under-curve) of ROC for K^trans (%change) is greater than that of ROC for k_ep (%change), suggesting that K^trans have higher predictive value for stroma-directed drug.
Notably, using the group-AIF, the M-T model revealed a moderate but significant increase of V_p after PP treatment (Fig 4A). Analysis of CD31 (endothelial marker) stained sections suggest that “vascular lumen area” metric exhibits an average of 48% increase after PP versus VEH treatment, although significance is not reached due to small sample size. Micrographs from one PP and one VEH treated mouse are shown in Fig 4B-4C.
DISCUSSION: The group-AIF combined with SS-model (not with T or M-T model) can detect response to PP, while individual AIF can work with T or SS-model to detect such response. Our data suggest K^trans an optimal marker in sensitivity /specificity compared to k_ep.
CONCLUSION: This study provides rationale for choice of PK model and AIF approach which lead to quantitative DCE-MRI marker of optimal sensitivity and specificity for detecting PDA response to PEGPH20 treatment. Because imaging markers such as K^trans map can evaluate responses of the entire tumor, they may avoid the sampling error and morbidity associated with ultrasound-guided endoscopic biopsy employed in clinic trials of stromal drugs.

Acknowledgements

This study was partially supported by U24CA231858 (Penn Quantitative Imaging Resource for Pancreatic Cancer), R21CA198563 and R01CA211337.